The attached patch fixes a general problem using bonding with slaves
that have other than the default ETH_HLEN hard_header_len. For example,
the GIANFAR driver may push a control block on the front of the packet
(instead of in the descriptors, where sane hardware expects it 8-). As
a result, the driver specifies hard_header_len is increased to handle
the control information.
When such an interface is enslaved by bonding, the kernel soon oopes or
panics, depending on where the next transmit comes from. This is
because skb_push fails with an underflow due to inadequate head room in
Get latest updates about Open Source Projects, Conferences and News.